Title: Comparison of Some Estimation Procedures in Surveys of Skewed Population Author: Maria Praxedes O. Reyes Degree: Master of Science in Statistics Date: March 2006 Abstract: The complexity in survey estimation is confounded by the presence of nonresponse apart from sampling and other nonsampling errors. This problem is further aggravated by skewness which is common among count variables and in establishment surveys focusing on labor characterization. Nonresponse and ineligible respondents can be addressed by sequential weight-adjustment techniques while resampling methods provide alternative procedures in estimating variances from complex surveys. When the population is skewed, with pervasive nonsampling error, model-based estimation may provide a superior alternative to design-based estimate. A generalized linear model and piecewise regression were used and an improvement in precision and accuracy of estimates is noted. On the other hand, the alternative weight-adjustment estimation method and estimation using bootstrap gives better results than the current design-based estimation method, jackknife estimation method and model-based estimation method. Both methods yield estimates that are close to the actual values. Simulation experiments were done to evaluate the effects of nonresponse to the different alternative estimation methods. These empirical evidences showed that performance of the estimators for the alternative weightadjustment method and bootstrap remains stable even at different levels of nonresponse.